Krivium, I fully agree with dBFS being for digital samples. However, there are a few other things that I cannot agree with:
That's applies only when samples are sent out of a processor. However, inside the processor, the datapath is (normally) much wider than 24 bits, often 28/32/40 bits. With these extra bits, it becomes possible to represent numbers that go above 0dBFS. For example, the largest 24-bit fixed point number is FFFFFF, but the highest 32-bit number is FFFFFFFF, about 256 times bigger.
Thus, not only can a processor represent numbers beyond 0dBFS, they can also perform calculations on them, that include comparison and gain/attenuation. The output from a processor is then scaled and rounded to 24-bits (final precision) of which all ones give 0dBFS.
Well, there are, in fact, a hundred standards instead of one, which is precisely why it's best to say that there is no standard. The only reason people still try to find a common ground is because, in the pro world, the show must go on irrespective of compatibility.
Yes, but if the amplifiers and speakers are rated high enough to take peaks, the limiter is then just a guard against wrong settings, unexpected problems etc.
krivium said:
That's applies only when samples are sent out of a processor. However, inside the processor, the datapath is (normally) much wider than 24 bits, often 28/32/40 bits. With these extra bits, it becomes possible to represent numbers that go above 0dBFS. For example, the largest 24-bit fixed point number is FFFFFF, but the highest 32-bit number is FFFFFFFF, about 256 times bigger.
Thus, not only can a processor represent numbers beyond 0dBFS, they can also perform calculations on them, that include comparison and gain/attenuation. The output from a processor is then scaled and rounded to 24-bits (final precision) of which all ones give 0dBFS.
krivium said:
Well, there are, in fact, a hundred standards instead of one, which is precisely why it's best to say that there is no standard. The only reason people still try to find a common ground is because, in the pro world, the show must go on irrespective of compatibility.
krivium said:
Yes, but if the amplifiers and speakers are rated high enough to take peaks, the limiter is then just a guard against wrong settings, unexpected problems etc.
Krivium, your example tries to get the same SPL out of both amplifiers, that's why. If the final 3dB of an amplifier is not to be touched due to ISP/headroom, then the 64W amplifier is only good enough for 32W and the SPL that 32W provides. But you're driving it up to 64W, allowing the anomalies like ISP etc. to clip the amplifier.
Newvirus,
So we agree on dbfs. That is nice.
About digital signal and numbers of bits you are right about processing not about the rest of it ( and you mix different things).
Floating point allowed us to have a 'sliding scale' for treatment but once done the stream is sent to 24bit ( at most). There is no over in floating points because the end of scale ( odbfs) is scaled/slided. In no way it pass above 0dbfs.
In no way a processor will output something over 0dbfs. This is a hard fact and basic of digital understanding.
You can turn it either way you want this is how it is ( and tbh i don't see the point to argue about it as you just accepted what dbfs is in the first sentence of your post or the last sentence related to it).
The fact that once in analog there is artefact from reconstruction is in no way related to what happen inside the processor.
Sorry but you'll have to dig deeper on the subject or approach things in another way as there is misconception in your view.
About standard in proworld you have here again a fantasy approach to this: there is not hundreds of standards and no one as value.... there is a myriad of standard because we have 120 years history and technology evolutions.
Most standards comes from telecomunication ( phone) dbm and its 600ohms rated line, dbV which was the standard used for vinyl cutters and was then defacto accepted as the 'domestic' standard, dbu once the industry shifted to bridged line rather than power oriented...
I could go on and on and send you the AES norm attached to them but... i've got other things to do.
All this makes me think you should here again go deeper in your search. And i always find funny when enthousiast/amateur think they know things which was defined by an industry ( pro) because they don't understand the principle which leaded to the solution used.
Finally your last post just show you didn't understand the principle of headroom. I won't explain it and let you here again makes your own search about this.
I'm sorry this turn out this way but i don't want to have the role of the teacher to you ( please note this is not something i dislike as it was one of my job in a school of audio engineer) as like you i am here to learn things thanks to exchange with other members.
I don't like to give an image of a 'know it all' kind of guy, but all we talk about was my day to day basis some years ago so... i understand you question my answers ( and welcome it!) but please if you are not in agreement don't answer things like 'norm are meaningless', it just stop the will to help from my side.
So we agree on dbfs. That is nice.
About digital signal and numbers of bits you are right about processing not about the rest of it ( and you mix different things).
Floating point allowed us to have a 'sliding scale' for treatment but once done the stream is sent to 24bit ( at most). There is no over in floating points because the end of scale ( odbfs) is scaled/slided. In no way it pass above 0dbfs.
In no way a processor will output something over 0dbfs. This is a hard fact and basic of digital understanding.
You can turn it either way you want this is how it is ( and tbh i don't see the point to argue about it as you just accepted what dbfs is in the first sentence of your post or the last sentence related to it).
The fact that once in analog there is artefact from reconstruction is in no way related to what happen inside the processor.
Sorry but you'll have to dig deeper on the subject or approach things in another way as there is misconception in your view.
About standard in proworld you have here again a fantasy approach to this: there is not hundreds of standards and no one as value.... there is a myriad of standard because we have 120 years history and technology evolutions.
Most standards comes from telecomunication ( phone) dbm and its 600ohms rated line, dbV which was the standard used for vinyl cutters and was then defacto accepted as the 'domestic' standard, dbu once the industry shifted to bridged line rather than power oriented...
I could go on and on and send you the AES norm attached to them but... i've got other things to do.
All this makes me think you should here again go deeper in your search. And i always find funny when enthousiast/amateur think they know things which was defined by an industry ( pro) because they don't understand the principle which leaded to the solution used.
Finally your last post just show you didn't understand the principle of headroom. I won't explain it and let you here again makes your own search about this.
I'm sorry this turn out this way but i don't want to have the role of the teacher to you ( please note this is not something i dislike as it was one of my job in a school of audio engineer) as like you i am here to learn things thanks to exchange with other members.
I don't like to give an image of a 'know it all' kind of guy, but all we talk about was my day to day basis some years ago so... i understand you question my answers ( and welcome it!) but please if you are not in agreement don't answer things like 'norm are meaningless', it just stop the will to help from my side.
Last edited:
Like krivium has been saying, dBFS is not valid at all for amplifier/speaker limiting.
All dBFS limiting does is limit the line-level voltage coming out of the line-level device.
It cannot know what are the amps' input sensitivity, gain, and power.
It cannot know what are the drivers thermal (program/RMS/etc) and peak (excursion) limits.
Amp limiting requires a mapping of the line-device output, into those amp and speaker parameters.
There is no escaping the voltage math necessary to do so.
Whether you calculate directing with voltage, or use gain in dBu or dBV doesn't matter...they're all voltage calcs.
The rest of limiting is time of attack and release, and is frequency dependent for both RMS/program and peak limiting. And requires different attack and release times for both.
Pages 1 & 7 of the MC2 limiting paper do show the same attack and release recommendations for program and peak limiting.
It appears he reason they do is because MC2 is assuming only a 3dB spread between program and peak voltages. (this can be seen top of page 7.)
That 3dB spread is very tight and makes the program limiter no more than an early bird peak limiter, imo ........hence the same attack and release times.
6dB or more is a far more typical spread between RMS and peak limiter settings.
Lower voltage settings than AES rating imply, and longer attack and release times are the norm for RMS/program limiting.
Any sustained voltage close to RMS/program will cause at least 3dB attention somewhere in the drivers passband, and often more.
The MC2 Program attack and release times are simply too fast ime.....and will like create a pumping sound when being on and off activated.
For example, a RMS/thermal sub limiter should take a number of seconds to attack, and a multiple of that attack release.
The big thing though, is dBFS simply cannot be taken alone to have meaning on amp/speaker limiting.
And it's not a matter of PA vs home....the science is the same....it's just a matter of when is the science is needed.
ps....also like krivium, don't mean to sound like a teacher...just trying to cut throught the chase a little and hopefully help out
Last edited:
krivium said:
I do not understand where I said that a processor could provide levels over 0dBFS. All I said was that a processor's wider internal word-length allows it to perform limiting on sample values above 0dBFS, so that its output is restricted to be within the correct range i.e. 0dBFS. I just don't understand how that is a misconception. The limiter is just a arithmetic and logical module inside the processor, just like biquads and scalers!
I had already said in #22 that I can say dB (or dB something else), if you have an issue with me saying 'dBFS', which has multiple standards associated with it, as already given on the wikipedia page cited by you earlier.
krivium said:
I never said that and I live by several norms everyday. However, a norm makes sense only if it's uniform everywhere, like g = 9.8m/s2.
mark100 said:
True that it doesn't know all that, which is why we're there to set it up. The machine's just a fool, it's the man behind it that makes it work correctly.
And what's wrong in setting a limiter in dB relative to amplifier output? After all, if the user already knows how many dBs it takes for the speaker to reach peak excursion, then what is the issue with simply dialing that value into the limiter settings? I just don't seem to get that part.
Last edited:
What I don't understand about this paper is, why they just don't just suggest brickwall limiter while they are talking about digital limiter.
A brick wall limiter has much less distortion (psychoacoustically) than any positive attack time limiter, because we can't really hear distortion right before the large transient. It's way better to reduce the level (modulate) before the transient than after transient theoretically.
I'm not a sound reinforcement guy, so probably I misunderstand something here...
Well, they do suggest something like that in the end, on the last page of the article:
"Only the release time may be adjusted for the peak limiters, as attack time is always set to “zeroovershoot” and so cannot be changed. The release time may be set to “slow”, “medium” or “fast” —..."
Thanks for confirming that.
"Only the release time may be adjusted for the peak limiters, as attack time is always set to “zeroovershoot” and so cannot be changed. The release time may be set to “slow”, “medium” or “fast” —..."
mark100 said:
Thanks for confirming that.
Last edited:
mark100 said:
Yes, they are, but if it's in dB, it remains possible to compare between different settings/methods, as it's relative to full power. After all, every amplifier starts clipping at 0dB!
Some manufacturers use kW, while some others use hp, but that's rarely an issue, as you could convert between these anytime. Plus, if you are to compare a car with a bus (entirely different engines), you could still talk in relative terms, such as full power, half speed etc. For limiters, you could say something like this:
"I usually set my limiters to 0dB, but my friend uses a -3dB setting, as he is of the strong opinion that a 3dB headroom is necessary for accommodating the inter-sample peaks that occur during the D/A conversion process."
Now, note that the above sentence does not carry any information about amplifiers, speakers, SPL or listening distances of the subject (or his friend), but still makes perfect sense. However, for limiter settings in dBu, it could become:
"I usually set my limiters at +4dBu beyond which my "amplifier X" goes into hard clipping, but my friend, even with his bigger "Y amplifier", uses a +1dBu limit to protect his expensive "Z speakers" from excursion damage."
Now, the sentence just became more complicated, as the different speaker and amplifier models were brought into the picture.
However, like Mark said, it's the responsibility of each person to work out the correct thresholds for their respective systems in units required by their processors.
I hope the above example would help Mark and Krivium (and anyone else reading) understand why I talk decibels. However, I'm sure that someone would disagree as the decibels don't let them brag about their elaborate and expensive equipment!!
Last edited:
Hi newvirus2008, hope my posts have been helping...seems so, like we are converging on some understandings.
Ok.
Line level gear clips at 0dB......if by 0dB you mean 0dBFS for the digital signal.
But what about the analogue output of the line level device when at 0dBFS...what is its voltage, expressed in dB ?
You can't express the analog it in dB...it must be expressed in dBu or dBV, which are simply voltages. And even line devices analog outputs vary a lot.
Likewise, you can't express an amplifier's output in dB....
They don't clip at 0dB.
They clip at some voltage, which can be expressed in dBu if you like, and can range from only several volts to hundreds of volts.
And the way amps clip is vs time.
They have a maximum rail voltage, that for even instantaneous peaks, cannot be exceeded and will cause clipping.
But that maximum rail voltage can only be sustained for a brief period.
There is a curve for every amp, as to how long it can sustain what voltage level.
10-20ms is a fairly common burst time range that some manufactures use to define peak wattage output. Bottom feeder specs may even use something as short as 2ms.
Point of all the above, is to illustrate how deep the subject of amp limiting is vs simple brickwall type limiting.
A well implemented line level limiter, might well take into account not only speaker protection, but amplifier output limitation vs time.
Better than that though, is just have an amp that never clips into the load its driving, and all the line-level limiter does is limit RMS and peak, according to the thermal and excursion capabilities, without regard for amp limitations.
Better than that still, is don't use a line-level limiter for speaker protection.
A line-level limiter is a predictive limiter, predicting the amp's output voltage.
The very best is when the amp itself monitors itself, and limits according to the voltage settings put into it. (Becoming pretty standard with pro-amps)
And last i guess, i always like to stop and ask...what am i limiting and why?
If i wanted a limiter just to keep the amp from clipping, it would have to be because my line-level limiting sounds better than my amp clipping.
And that begs i understand the amps voltage vs time delievery curve if i want the amps maximum headroom.
(Which means i need a bigger amp to avoid all that crappola haha)
If I want speaker protection, i'm simply forced to do the math for predictive line-level limiting...or use an advanced amp with internal limiting, where i set RMS and peak voltages, and their corresponding attack and release times, simply off the speaker specifications.
Dear Mark,
In the earlier posts, I was taking 0dBFS to be the amplifier's maximum input (clipping) voltage, to which krivium objected, citing intersample peaks of 3dB (see quote below). Since then, I'm taking the amplifier's maximum input (clipping) voltage to be 0dB, the level that provides full power.
Since post #38, 0dBFS is 6dB lower, as recommended by krivium.
Digital:
❚❚❚❚❚❚❚❚❚❚❚❚❚❚❚❚❚❚❚❚❚❚❚❚❚❚❚ 0dBFS = -6dB on amp input, makes 1/4th of max amp. power
Analogue with ISP:
❚❚❚❚❚❚❚❚❚❚❚❚❚❚❚❚❚❚❚❚❚❚❚❚❚❚❚❚❚❚❚❚❚❚ -3dB on amp input, half power at amp output
Analogue + ISP + Krivium headroom:
❚❚❚❚❚❚❚❚❚❚❚❚❚❚❚❚❚❚❚❚❚❚❚❚❚❚❚❚❚❚❚❚❚❚❚❚❚❚❚❚❚❚❚ 0dB on amp input, full power point, onset of clipping (only if driven to)
The dBs are relative to the input clipping voltage of amplifier. Thus, the amplifier clips at 0dB level, whatever voltage that returns, could be a few volts to a few hundred volts....like you said. Now, this assumes that the amplifier output is directly proportional to its input, which is reasonable, especially with the "very linear" amplifiers of today.
No, no, I'm referring to digital limiting inside the processor, after the biquads and before the D/A conversion stage, not the ones within the amplifier itself. This was why I said earlier that it is indeed possible to limit 0dBFS+ samples to 0dBFS if they were to happen for some reason (EQ, wrong settings etc.).
The purpose of the limiting is also not thermal (VC) protection, it's the prevention of audible clipping through appropriate processing in the digital domain. The calculations for RMS limiting (had I required) were already included in the user manual (page 13, 7.4.1) for the DSC-260 that I mentioned I was reading in post#1.
Please note that I'm mainly after the time-constants, not exactly the thresholds. But then, I decided to look into the thresholds in case they affected the time constants (for peak limiting). Hope this clears things.
Regards.
In the earlier posts, I was taking 0dBFS to be the amplifier's maximum input (clipping) voltage, to which krivium objected, citing intersample peaks of 3dB (see quote below). Since then, I'm taking the amplifier's maximum input (clipping) voltage to be 0dB, the level that provides full power.
krivium said:
Since post #38, 0dBFS is 6dB lower, as recommended by krivium.
Digital:
❚❚❚❚❚❚❚❚❚❚❚❚❚❚❚❚❚❚❚❚❚❚❚❚❚❚❚ 0dBFS = -6dB on amp input, makes 1/4th of max amp. power
Analogue with ISP:
❚❚❚❚❚❚❚❚❚❚❚❚❚❚❚❚❚❚❚❚❚❚❚❚❚❚❚❚❚❚❚❚❚❚ -3dB on amp input, half power at amp output
Analogue + ISP + Krivium headroom:
❚❚❚❚❚❚❚❚❚❚❚❚❚❚❚❚❚❚❚❚❚❚❚❚❚❚❚❚❚❚❚❚❚❚❚❚❚❚❚❚❚❚❚ 0dB on amp input, full power point, onset of clipping (only if driven to)
The dBs are relative to the input clipping voltage of amplifier. Thus, the amplifier clips at 0dB level, whatever voltage that returns, could be a few volts to a few hundred volts....like you said. Now, this assumes that the amplifier output is directly proportional to its input, which is reasonable, especially with the "very linear" amplifiers of today.
mark100 said:
No, no, I'm referring to digital limiting inside the processor, after the biquads and before the D/A conversion stage, not the ones within the amplifier itself. This was why I said earlier that it is indeed possible to limit 0dBFS+ samples to 0dBFS if they were to happen for some reason (EQ, wrong settings etc.).
The purpose of the limiting is also not thermal (VC) protection, it's the prevention of audible clipping through appropriate processing in the digital domain. The calculations for RMS limiting (had I required) were already included in the user manual (page 13, 7.4.1) for the DSC-260 that I mentioned I was reading in post#1.
Please note that I'm mainly after the time-constants, not exactly the thresholds. But then, I decided to look into the thresholds in case they affected the time constants (for peak limiting). Hope this clears things.
Regards.
Last edited:
Ok, i think what you are saying is that you are only concerned with digital clipping, and have been all along...in which case it hasn't ever made sense to extend clipping to the analog domain, for any device, either the processor itself or an amplifier.
That being the case, digital only....it seems simple to me... just don't let it clip....gain structure solves that, doesn't it?
Why would you need a digital clip limiter if gain structure is right? F
My apologies for being dense about digital limiting only....but when that digital limiting was talked about, extended to amps etc, i trust everything i said holds.
That being the case, digital only....it seems simple to me... just don't let it clip....gain structure solves that, doesn't it?
Why would you need a digital clip limiter if gain structure is right? F
My apologies for being dense about digital limiting only....but when that digital limiting was talked about, extended to amps etc, i trust everything i said holds.
Yes, I was mostly referring to the clipping inside a processor, with which the "don't clip" option is just not there since there're parametric EQs, shelves and gains for speaker compensation/crossover and it is definitely possible for a higher gain to be dialled in by mistake, or a wrong speaker setting to be loaded from the factory presets.
However, I was able to learn few useful things about RMS power limiting as well, thank you very much for that.
Now I guess the attack/release rates are equal to the gain/attenuation differences divided by the time-constants. That is for example, with a 1kHz crossover, 0dBFS threshold and a worst-case peak of +6dBFS, the LF attack rate would be (6-0)dB/1ms = 6dB/ms, with a one-tenth release rate (10x release time) of 0.6dB/ms, is that correct?
However, I was able to learn few useful things about RMS power limiting as well, thank you very much for that.
Now I guess the attack/release rates are equal to the gain/attenuation differences divided by the time-constants. That is for example, with a 1kHz crossover, 0dBFS threshold and a worst-case peak of +6dBFS, the LF attack rate would be (6-0)dB/1ms = 6dB/ms, with a one-tenth release rate (10x release time) of 0.6dB/ms, is that correct?
Glad to have helped some, even if not on your focus
I guess i feel like digital peak is digital peak, and doesn't need any time-constants. I still think it's a matter of gain structure and simply not allowing clipping.
My processor is an open architecture type, that allows for processing blocks such as gain control, xovers, EQ's, limiters, etc, to be routed any way desired, and in any order.
Also allows any number of peak or RMS meters to be put anywhere to monitor each block, if desired.
In the digital block flow, if I had any concerns about clipping with some particular block, I would place a peak meter on the block, and just insert a gain control if needed. But i'd never put a limiter in solve what i think is a gain structure problem. hope that made sense...
I can't see why a time constant would ever come into play for digital peak limiting....
Maybe it does in studio land...again, not my area of expertise....
Well, these days all fancy processing that used to be "pro-only", like RTA speaker compensation, biquads, active crossover, dynamic EQ, DRC, limiting etc. are all available on single-chip amplifier ICs, even for the most unexpected itsy-bitsy-power applications like mobile, IOT, home-theatre-in-box (HTIB), home automation etc.
Therefore, I hope the information in this thread would also be of some help to those who plan to use the limiting features on such chips, especially in multi-way systems.
Therefore, I hope the information in this thread would also be of some help to those who plan to use the limiting features on such chips, especially in multi-way systems.